This paper describe a view based localization method for mobile robot. Our proposal system use ground image as landmark to estimate of robot position. Using ground image as a landmark has several advantages that have many occasions to observe a landmark in crowd because of less occlusion and easy to create a map because of the 2-D plane. Our proposal system create a ground map by overlaying the ground images and estimate the robot position by maching feature points between the observed images and the ground map images. We realize a highly accurate self position estimate by resolving the problem of feature points matching such as feature points error and image saturated by sunlight.